PCN-with-Global-Recurrent-Processing/prednet.py at master · libilab/PCN-with-Global-Recurrent-Processing · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
'''PredNet in PyTorch.'''

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable


# Feedforeward module
class FFconv2d(nn.Module):
    def __init__(self, inchan, outchan, downsample=False):
        super().__init__()
        self.conv2d = nn.Conv2d(inchan, outchan, kernel_size=3, stride=1, padding=1, bias=False)
        self.downsample = downsample
        if self.downsample:
            self.Downsample = nn.MaxPool2d(kernel_size=2, stride=2)

    def forward(self, x):
        x = self.conv2d(x)
        if self.downsample:
            x = self.Downsample(x)
        return x


# Feedback module
class FBconv2d(nn.Module):
    def __init__(self, inchan, outchan, upsample=False):
        super().__init__()
        self.convtranspose2d = nn.ConvTranspose2d(inchan, outchan, kernel_size=3, stride=1, padding=1, bias=False)
        self.upsample = upsample
        if self.upsample:
            self.Upsample = nn.Upsample(scale_factor=2, mode='bilinear')

    def forward(self, x):
        if self.upsample:
            x = self.Upsample(x)
        x = self.convtranspose2d(x)
        return x


# FFconv2d and FBconv2d that share weights
class Conv2d(nn.Module):
    def __init__(self, inchan, outchan, sample=False):
        super().__init__()
        self.kernel_size = 3
        self.weights = nn.init.xavier_normal(torch.Tensor(outchan,inchan,self.kernel_size,self.kernel_size))
        self.weights = nn.Parameter(self.weights, requires_grad=True)
        self.sample = sample
        if self.sample:
            self.Downsample = nn.MaxPool2d(kernel_size=2, stride=2)
            self.Upsample = nn.Upsample(scale_factor=2, mode='bilinear')

    def forward(self, x, feedforward=True):
        if feedforward:
            x = F.conv2d(x, self.weights, stride=1, padding=1)
            if self.sample:
                x = self.Downsample(x)
        else:
            if self.sample:
                x = self.Upsample(x)
            x = F.conv_transpose2d(x, self.weights, stride=1, padding=1)
        return x

# PredNet
class PredNet(nn.Module):

    def __init__(self, num_classes=10, cls=3):
        super().__init__()
        ics = [3,  64, 64,  128, 128, 256, 256, 256] # input chanels
        ocs = [64, 64, 128, 128, 256, 256, 256, 256] # output chanels
        sps = [False, False, True, False, True, False, False, False] # downsample flag
        self.cls = cls # num of circles
        self.nlays = len(ics) #number of layers

        # Feedforward layers
        self.FFconv = nn.ModuleList([FFconv2d(ics[i],ocs[i],downsample=sps[i]) for i in range(self.nlays)])
        # Feedback layers
        if cls > 0:
            self.FBconv = nn.ModuleList([FBconv2d(ocs[i],ics[i],upsample=sps[i]) for i in range(self.nlays)])

        # Update rate
        self.a0 = nn.ParameterList([nn.Parameter(torch.zeros(1,ics[i],1,1)+0.5) for i in range(1,self.nlays)])
        self.b0 = nn.ParameterList([nn.Parameter(torch.zeros(1,ocs[i],1,1)+1.0) for i in range(self.nlays)])

        # Linear layer
        self.linear = nn.Linear(ocs[-1], num_classes)

    def forward(self, x):

        # Feedforward
        xr = [F.relu(self.FFconv[0](x))]
        for i in range(1,self.nlays):
            xr.append(F.relu(self.FFconv[i](xr[i-1])))

        # Dynamic process
        for t in range(self.cls):

            # Feedback prediction
            xp = []
            for i in range(self.nlays-1,0,-1):
                xp = [self.FBconv[i](xr[i])] + xp
                a0 = F.relu(self.a0[i-1]).expand_as(xr[i-1])
                xr[i-1] = F.relu(xp[0]*a0 + xr[i-1]*(1-a0))

            # Feedforward prediction error
            b0 = F.relu(self.b0[0]).expand_as(xr[0])
            xr[0] = F.relu(self.FFconv[0](x-self.FBconv[0](xr[0]))*b0 + xr[0])
            for i in range(1, self.nlays):
                b0 = F.relu(self.b0[i]).expand_as(xr[i])
                xr[i] = F.relu(self.FFconv[i](xr[i-1]-xp[i-1])*b0 + xr[i])

        # classifier
        out = F.avg_pool2d(xr[-1], xr[-1].size(-1))
        out = out.view(out.size(0), -1)
        out = self.linear(out)

        return out


# PredNet
class PredNetTied(nn.Module):
    def __init__(self, num_classes=10, cls=3):
        super().__init__()
        ics = [3,  64, 64,  128, 128, 256, 256, 256] # input chanels
        ocs = [64, 64, 128, 128, 256, 256, 256, 256] # output chanels
        sps = [False, False, True, False, True, False, False, False] # downsample flag
        self.cls = cls # num of circles
        self.nlays = len(ics) # num of circles

        # Convolutional layers
        self.conv = nn.ModuleList([Conv2d(ics[i],ocs[i],sample=sps[i]) for i in range(self.nlays)])

        # Update rate
        self.a0 = nn.ParameterList([nn.Parameter(torch.zeros(1,ics[i],1,1)+0.5) for i in range(1,self.nlays)])
        self.b0 = nn.ParameterList([nn.Parameter(torch.zeros(1,ocs[i],1,1)+1.0) for i in range(self.nlays)])

        # Linear layer
        self.linear = nn.Linear(ocs[-1], num_classes)

    def forward(self, x):

        # Feedforward
        xr = [F.relu(self.conv[0](x))]
        for i in range(1,self.nlays):
            xr.append(F.relu(self.conv[i](xr[i-1])))

        # Dynamic process
        for t in range(self.cls):

            # Feedback prediction
            xp = []
            for i in range(self.nlays-1,0,-1):
                xp = [self.conv[i](xr[i],feedforward=False)] + xp
                a = F.relu(self.a0[i-1]).expand_as(xr[i-1])
                xr[i-1] = F.relu(xp[0]*a + xr[i-1]*(1-a))

            # Feedforward prediction error
            b = F.relu(self.b0[0]).expand_as(xr[0])
            xr[0] = F.relu(self.conv[0](x - self.conv[0](xr[0],feedforward=False))*b + xr[0])
            for i in range(1, self.nlays):
                b = F.relu(self.b0[i]).expand_as(xr[i])
                xr[i] = F.relu(self.conv[i](xr[i-1]-xp[i-1])*b + xr[i])

        # classifier
        out = F.avg_pool2d(xr[-1], xr[-1].size(-1))
        out = out.view(out.size(0), -1)
        out = self.linear(out)
        return out